Research On Quasi-circular Color Image Segmentation Technology

In the fields of machine vision and image analysis, quasi-circular image segmentation and recognition is an important research direction. Quasi-circular, as the name suggests. is shaped nearly like a circle. Quasi-circular objects exist everywhere in real life, such as:industrial parts, animal cells, fruits, seeds, log cross sections, steel bars, all kinds of balls, metal coins and so on. And quasi-circular objects usually have such characteristics:irregular shape, incomplete border, unequal size, uneven edge, and even severe adhesion, therefore there are some difficulties in practical application. Quasi-circular image segmentation and recognition is a hotspot and its research is of great importance both in theory and practice.According to the characteristics of quasi-circular images, a quasi-circular image segmentation and recognition algorithm based on distance image is put forward in this paper. In the hardware, CCD camera, digital image acquisition card and PC are utilized to realize image acquisition, segmentation and recognition in real time. As to the software aspect, the algorithm is achieved with C/C++programming language and Visual C++6.0software development tool on Windows XP operating system. The advantages of multi-threading technology are taken full advantage of and the storage management is optimized to enable the system more stable and reliable. For quasi-circular image segmentation and recognition, firstly, the color image captured by CCD camera is processed with pretreatment methods:and then, the high quality binary image obtained is processed with distance transform based on boundary-stripped algorithm:at last, the distance image is considered as a contour map in topography and the quasi-circular objects can be detected by searching the maximum pixel point or area. By means of setting a reasonable threshold and the zero setting area of imitate circles, the real-time property is improved to a certain extent. The simulation results indicate that the method is simple. has a high recognition rate and can identify and count the adhesion image rapidly and effectively.